Vision-based Location And Defect Recognition Technology For Without Buffing Welds

With the further improvement of port machinery manufacturing requirements,welding quality has become an important link to ensure production safety and product safety.The domestic over-reliance on manpower for welding seam quality inspection has caused problems such as low production speed and high production risks.The use of robots to complete weld defect quality inspection is one of the keys to solving the above-mentioned problems.Aiming at this difficulty,this paper studies the robot system for welding seam positioning and defect recognition based on vision.There are three main links in the whole robot system: Image acquisition system,image processing system of weld location and defect recognition,robot positioning system with industrial camera.The software and hardware equipment in these three links are interconnected,and the data information is passed down along the first link,so as to complete the goal of robot weld positioning and defect identification.In order to study the goal of welding seam location and defect recognition in large components,the welded metal workpiece with weld seam is used as the research object to carry out work.In view of the above three links and research goals,the work carried out includes the following contents:(1)In order to make the metal workpiece with welded seam simulate the lighting conditions in the large-component factory environment,a set of optical platform was built to collect images of the workpiece.First,the effect of different kinds of light sources on the workpiece is studied.The experiment shows that the dome light has a good effect on the workpiece.The strip light can simulate the work piece being interfered by strong light.Combining the lighting effect of the workpiece,build a combined light lighting system,the industrial camera completes the image collection,and the computer collects the image data.Thus,the acquisition system of the workpiece image is completed.(2)In order to solve the problem of accurate positioning of weld and accurate recognition of defects in image processing,two kinds of image processing ideas are tried.One is the traditional image processing scheme based on three methods of threshold,edge extraction and local gray mean value.The other is to use the deep learning method based on FPN network structure,apply the target detection function of the network to locate the weld,and apply the semantic segmentation function to detect the weld defects.After network optimization,the accuracy of weld location and defect identification is 95% and 91.8%,respectively.Comparing the experimental results of the two methods,it is found that the deep learning method has higher positioning and recognition accuracy and better stability than the traditional image processing method.(3)In order to automatically guide the robot tool to the position of the weld and the workpiece with weld defects,the industrial camera and the robot are installed together,and the industrial camera is used to provide the visual function.Firstly,the coordinate position of the weld in the image coordinate system is calculated;according to the mapping relationship,the coordinate points are transformed into the tool coordinates defined by KUKA robot system to realize the positioning of the target area.By using QT,workvisual and other programming tools,the PC end program and robot end program are written,and the communication between PC end and robot end is realized.To sum up,this article solves the technical problems in the three major links of weld location and defect recognition,and completes the task of guiding the robot to the workpiece weld position and defect location by the vision system.It has a good reference for automatic defect detection.It has a certain practical value.